Acoustic-Inertial Forward-Scan Sonar Simultaneous Localization and Mapping

Abstract: The increasing accessibility and versatility of forward-scan (FS) imaging sonars (also known as forward looking sonars or FLS) has spurred the interest of the robotics community seeking to solve the difficult problem of robotic perception in low-visibility underwater scenarios. Processing the incoming data from an imaging sonar is challenging, since it captures an acoustic 2D image of the 3D scene instead of providing straightforward range measurements like other sonar technologies do (e.g. multibeam sonar). Hence, complex postprocessing and sensor fusion techniques are required to extract useful information out of the sonar image. The present report details development, validation and implementation of an acoustic-inertial localization and mapping algorithm that processes sonar images captured by an FS sonar and inertial measurements to solve the simultaneous localization and mapping (SLAM) problem with an underwater sensor suite. A sonar odometry pose constraint is computed by detecting and matching features from two consecutive sonar images on a degeneracy-aware two-view bundle adjustment. The sonar odometry measurements are fused with preintegrated inertial measurements in a minimal pose-graph representation. The state-of-the-art iSAM2 (Incremental Smoothing and Mapping) solver is used to allow for real-time localization. A Python simulator was developed to evaluate the performance of the two-view bundle adjustment algorithm. Results are presented and discussed from both computer simulations in Gazebo using the Robot Operating System (ROS) and from real-world tests in a controlled environment with an in-house developed sensor suite. Sonar image degeneracies, sensor drift, and computation complexity, proved to be hard to tackle, reducing the performance and robustness of the current implementation of the SLAM solution. However, the current work will serve as a stepping stone for for future work and collaboration in underwater localization and mapping using FS sonars.